Modeling Gross Primary Production in Maize and Soybean Using Four Parameters: Light Quality, Temperature, Water Stress, and Phenology

Friday, 19 December 2014
Andrew Suyker1, Anthony L Nguy-Robertson1, Xiangming Xiao2 and Taylor Thomas1, (1)University of Nebraska Lincoln, Lincoln, NE, United States, (2)University of Oklahoma, Norman, OK, United States
Light use efficiency (LUE) models are used to estimate gross and net primary production (GPP, NPP). Earlier approaches have used one or more of the following factors impacting LUE to model GPP: (i) light climate, (ii) temperature, (iii) water stress, and (iv) phenology. In this study we seek to incorporate all four inputs as scalars for up- or down-regulating LUE in a daily GPP model. Traditional methods using satellite data are limited by cloudy conditions and revisit times. They require the use of interpolation to achieve daily estimates of GPP. Some parameters can vary greatly within a few days (i.e. light climate, temperature) and thus, reduce the accuracy of interpolations between scenes. Alternatively, methods that can combine spatially interpolated gridded meteorological and satellite data for parameters that change over the course of days to weeks (i.e. phenology) provide the best opportunity for a continuous daily estimate of GPP. This study seeks to develop the framework for such a model. Three Nebraska AmeriFlux sites between 2001 and 2012 (maize: 26 field-years; soybean: 10 field-years) were used to develop and validate a daily GPP model based on ground measurements of incoming photosynthetically active radiation, temperature, vapor pressure deficit, and leaf area index. This model was calibrated using eddy covariance data from 2001 to 2008 (RMSE = 2.2 g C m-2 d-1; MNB = 4.7%) and validated with 2009 to 2012 data (RMSE = 2.6 g C m-2 d-1; MNB = 1.7%). Modeled GPP was generally within 10% of measured growing season totals in each year from 2009 to 2012. Cumulatively, over the same four years, the sum of error and the sum of absolute error between the measured and modeled GPP, which provide measures of long-term bias, was ±5% and 2 to 9%, respectively, among the three sites. The inclusion of AmeriFlux cropland sites in Iowa, Illinois, and Minnesota will support this approach and provide additional validation sites for a daily GPP model using gridded meteorological and remote sensing products.